加入新推出的
Discord 社区,展开实时讨论,获得同行支持,并直接与 Meridian 团队互动!
常见问题解答
使用集合让一切井井有条
根据您的偏好保存内容并对其进行分类。
该库是否免费?
是,您可以免费使用该库,并且我们在 GitHub 上开放了源代码,供所有人使用。
我们是否必须与 Google 共享数据才能使用该库?
Google 无法访问您的输入数据、模型或结果(通过我们的 MMM Data Platform 提供的 Google 媒体数据除外)。Google 只能访问您向 Google MMM Data Platform 请求的数据。但 Google 并不知道您的模型中是否真的包含了这些数据。除非您选择与 Google 代表共享数据,否则您的实际模型输入和输出是完全私密的。
从 LightweightMMM 迁移
如果我目前是 LightweightMMM 用户,是否需要执行额外的操作才能为 Meridian 构建数据输入?
若要充分利用新的 Meridian 创新功能,您需要添加更多数据维度,例如:
-
覆盖面和频次
-
实验
-
Google 搜索查询量 (GQV)
在没有这些维度的情况下,您仍可以运行 Meridian,但会错失新的创新功能。如需了解详情,请参阅从 LightweightMMM 迁移。
数据收集和清理
我能否在 MMM Data Platform 界面中同时收集所有数据类型(效果、YouTube 覆盖面和频次、Google 搜索查询量)?
效果数据与 YouTube 覆盖面和频次数据必须分开请求。如需详细了解请求工作流程,请参阅 MMM Data Platform 体验电子邮件中附带的用户指南。
我可以请求哪些 GQV 数据?
Google 搜索查询量,输出包括:
-
QueryLabel - 品牌或宽泛
-
ReportDate
-
TimeGranularity(您可以请求 Daily、Weekly_Sunday 或 Weekly_Monday 数据)。
-
GeoCriteriaId
-
GeoName
-
GeoType
-
IndexedQueryVolume - 所有搜索查询量数据均已经过指数化处理。系统不会提供搜索查询量的原始数据。
是否可以将 GQV 方法应用于非 Google 搜索数据?
来自非 Google 搜索引擎的自然搜索查询量往往无法获得。如需了解一些替代方案,请参阅了解作为搜索广告混杂因素的搜索查询量。
基于模型的分析
对于给定的媒体渠道,如何设置与不同时间段关联的不同先验?
使用 roi_calibration_period
实参最有可能做到这一点。根据 MMM 校准白皮书的第 3.4 部分,我们建议您计算实验按支出加权的平均投资回报率,并传递 roi_calibration_period
以匹配实验的四个季度。如果实验的标准误差相差较大,您可能需要进一步对实验进行相应的加权处理。如需了解详情,请参阅设置投资回报率校准周期。
能否为结值设置时间先验?
Meridian 不支持对结值设置时变先验。
能否在 Meridian 中衡量渠道之间的协同效应?
Meridian 不支持此类分析。
能否通过 Meridian 获得投资回报率随时间变化的数据?
您可以查看每个媒体渠道在一段时间内的增量结果,从而计算投资回报率:
-
获取
Analyzer().incremental_outcome()
中估计的增量结果。
-
使用
selected_times
选项选择相关的周。
-
除以这些周的支出。这样就得出了投资回报率,并能更准确地反映各个时间段的情况。
重要提示:在跟踪投资回报率随时间的变化时,要考虑到尽管模型中的系数不随时间变化,投资回报率仍会随时间变化,因为它依赖于可能随时间变化的其他因素。例如,Hill 曲线模拟了媒体投放的非线性递减回报,因此给定时间的媒体投放量会影响投资回报率。此外,不同地理位置的媒体分配会随着时间的推移而变化,并产生不同的效果,媒体投放费用也会随时间的推移而变化。
解读和优化
能否根据设定的出价目标值来衡量出价策略的投资回报率?
Google 的 MMM 数据 Feed 按广告系列提供出价策略类型(例如“尽可能提高转化次数”出价策略和“目标广告支出回报率”出价策略),但该 Feed 不包含出价目标值本身。若要查看这一特定维度,广告客户可以直接从 Google Ads 获取出价策略报告,也可以在其 Google 客户代表的帮助下量身定制数据解决方案。
如未另行说明,那么本页面中的内容已根据知识共享署名 4.0 许可获得了许可,并且代码示例已根据 Apache 2.0 许可获得了许可。有关详情,请参阅 Google 开发者网站政策。Java 是 Oracle 和/或其关联公司的注册商标。
最后更新时间 (UTC):2025-07-29。
[null,null,["最后更新时间 (UTC):2025-07-29。"],[[["\u003cp\u003eMeridian is an open-source MMM library currently in early access, best suited for advertisers with in-house MMM expertise and data.\u003c/p\u003e\n"],["\u003cp\u003eEarly access users are asked to provide feedback, participate in discussions on GitHub, and report issues.\u003c/p\u003e\n"],["\u003cp\u003eAccess is granted through an application form with limited spots, and invitations are sent quarterly.\u003c/p\u003e\n"],["\u003cp\u003eGoogle cannot see user data, model specifications, or results, unless data is requested from the Google MMM Data Platform.\u003c/p\u003e\n"],["\u003cp\u003eCurrent LightweightMMM users can migrate to Meridian but may need to add data dimensions to leverage new innovations.\u003c/p\u003e\n"]]],["The library is free and open-source on GitHub, with user data privacy maintained unless shared with Google representatives. Users can leverage new features by adding reach, frequency, experiment, and Google Query Volume (GQV) data. Performance and YouTube data are requested separately, while GQV includes indexed query data like Brand, ReportDate, and Geo details. Meridian does not support time-varying priors for knot values, measuring synergies between channels, or bid targets but enables calculating time-varying ROI by accessing incremental impact and dividing it by spend.\n"],null,["# FAQs\n\nGeneral product information\n---------------------------\n\n#### Is the library free?\n\n\nYes, the library is free to use and is open sourced on\n[GitHub](https://github.com/google/meridian) for anyone to use. \n\n#### Do we have to share our data with Google to use the library?\n\n\nGoogle won't have access to your input data, model, or results (apart from\nGoogle media data supplied through our MMM Data Platform). If you request data\nfrom the Google MMM Data Platform, that is the only data that Google has\naccess to. But Google won't know whether you actually include that data in\nyour model. Your actual model inputs and outputs are entirely private, unless\nyou choose to share it with your Google representatives.\n\nMigrating from LightweightMMM\n-----------------------------\n\n#### As a current LightweightMMM user, is extra work needed to build data input\nfor Meridian?\n\n\nTo take full advantage of the new Meridian innovations, you will need\nto add more data dimensions such as:\n\n- Reach and frequency\n- Experiments\n- Google Query Volume (GQV)\n\n\nYou can still run Meridian without these dimensions, although you will\nmiss out on the new innovations. For more information, see\n[Migrate from LightweightMMM](/meridian/docs/migrate).\n\nData collection and cleaning\n----------------------------\n\n#### Can I collect all data types simultaneously (performance, YouTube reach and\nfrequency, Google Query Volume) in the MMM Data Platform\ninterface?\n\n\nPerformance data and YouTube reach and frequency data must be requested\nseparately. The request workflow is detailed in the User Guide that is\nincluded with your MMM Data Platform access email. \n\n#### What is the scope of the GQV data that I can request?\n\n\nThe Google Query Volume, the output includes:\n\n- QueryLabel - Brand or generic\n- ReportDate\n- TimeGranularity (You can request Daily, Weekly_Sunday, or Weekly_Monday data.)\n- GeoCriteriaId\n- GeoName\n- GeoType\n- IndexedQueryVolume - All query volume data is indexed. Raw numbers aren't provided for Query Volume. \n\n#### Can I apply the GQV methodology for non-Google search data?\n\n\nOrganic query volume from non-Google search engines is often unavailable. Some\nalternative options are described in [Understanding query volume as a confounder for search ads](/meridian/docs/advanced-modeling/paid-search-modeling#understanding-query-volume-confounder).\n\nModeling\n--------\n\n#### For a given media lever, how can I set different priors associated with\ndifferent time periods?\n\n\nThe closest thing to this would be the `roi_calibration_period`\nargument. Based on section 3.4 of the [MMM calibration white paper](https://research.google/pubs/media-mix-model-calibration-with-bayesian-priors/), we suggest calculating a spend-weighted average ROI for the experiments\nand passing `roi_calibration_period` to match the four quarters\nof the experiments. If the experiments have very different standard errors,\nyou might want to further weight the experiments accordingly. For more\ninformation, see [Set the ROI calibration period](/meridian/docs/user-guide/configure-model#set-roi-calibration-period). \n\n#### Can I put a temporal prior for the knot values?\n\n\nMeridian does not support time varying priors for knot values. \n\n#### How can I get detailed decomposition information of the regression, such as\ngetting dataframes for the posterior draws?\n\n\nPosterior samples are in the `inference_data` object, and you can\nturn this array into any dataframe you need. To access the data samples using\nthe docstring, see [meridian.model.model.Meridian](/meridian/reference/api/meridian/model/model/Meridian). \n\n#### Can I measure synergies between channels in Meridian?\n\n\nMeridian doesn't support this kind of analysis. \n\n#### Is it possible to get a temporal read-out of ROI with Meridian?\n\n\nYou can access the incremental outcome of each media channel over time, and\ntherefore calculate ROI:\n\n1. Take the estimated incremental outcome, as found in [`Analyzer().incremental_outcome()`](/meridian/reference/api/meridian/analysis/analyzer/Analyzer#incremental_outcome).\n2. Use the `selected_times` option to choose the weeks of interest.\n3. Divide by spend over those weeks. This gives you the ROI and reflects the individual time period more accurately.\n\n\n**Important:** When tracking ROI over time, consider that even though the\ncoefficients in the model are not time-varying, the ROI can still change over\ntime because it is dependent on additional factors that might vary across\ntime. For example, the Hill curves model the non-linear, diminishing returns\nof media execution, and therefore the amount of media execution at a given time\ncan impact the ROI. Addtionally, media allocation can vary across\ngeos over time with different effectiveness and the cost of media\nexecution can vary across time.\n\nInterpretation and optimization\n-------------------------------\n\n#### Can I measure the ROI of bidding strategies based on the bid targets set?\n\n\nGoogle's MMM data feed provides bid strategy type (such as Maximize\nConversions and Target ROAS) by campaign, but the feed does not include the bid\ntarget itself. To view this specific dimension, advertisers can source bid\nstrategy reports directly from Google Ads or work with their Google Account\nRepresentative on a custom data solution."]]